Runner-cache dispatch falls back to the public cache when a Kura node goes unready

When a private Kura cache node stops being ready, CI builds and macOS runner builds no longer time out trying to download modules. The runner-cache dispatch now serves a node-port private endpoint only while a readiness heartbeat on the server is fresh (within ~120s). Once the heartbeat goes stale, dispatch returns nothing and the build falls back to the public Tuist cache instead of hitting a dead endpoint and timing out. Cluster-DNS private servers are unaffected, since their in-cluster Service already drops not-ready pods.